ChainSWE is a benchmark of sequential bug-fix tasks for evaluating coding agents on continuous software maintenance. Each row is one time-ordered sequence in one repository, with a shared base commit and Docker image; bugfixes contains its ordered tasks.
Dataset Card
By Qirui Jin, published under mit, revision a262e53112b5.
ChainSWE is a benchmark of sequential bug-fix tasks for evaluating coding agents on continuous software maintenance. Each row is one time-ordered sequence in one repository, with a shared base commit and Docker image; bug_fixes contains its ordered tasks.
The length-specific splits contain 376 sequences and 1504 bug-fixes.
| Split | Sequences | Bug-fix occurrences |
|---|---|---|
len_3 |
142 | 426 |
len_4 |
92 | 368 |
len_5 |
142 | 710 |
unique |
213 | 956 |
unique is a non-overlapping sequence union from the three length-specific splits, not additional data.
Evaluation
Evaluation is run with the SWE-Edit.
Details
- Repository
- C-lister/ChainSWE
- Publisher
- Qirui Jin
- Task category
- Text generation
- Tags
- code, software-engineering, bug-fixing
- Size category
- n<1K
- Languages
- en
- Revision
- a262e53112b59127d7f27b43de36e65121f554ea
- Last updated
- 2026-10-11
Files
6 files, 19.8 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| data/len_3.parquet | Data | 3.7 MB | e959bd4811e6 |
| data/len_4.parquet | Data | 3.7 MB | bf8e7d044e7b |
| data/len_5.parquet | Data | 4.4 MB | b0ef18f36827 |
| data/unique.parquet | Data | 8.0 MB | f7de0d66b64a |
| README.md | Documentation | 1.1 KB | — |
| .gitattributes | Repository | 1.7 KB | — |
License and Download
- License
- mit
- Access
- No access gate
Released by Qirui Jin through its official repository on Hugging Face. Read the license.